aacenc: add support for coding of IS spectral coefficients
authorRostislav Pehlivanov <atomnuker@gmail.com>
Thu, 2 Jul 2015 18:13:06 +0000 (19:13 +0100)
committerMichael Niedermayer <michaelni@gmx.at>
Sun, 5 Jul 2015 14:58:37 +0000 (16:58 +0200)
This commit adds support for the coding of intensity stereo spectral
coefficients. It also fixes the Mid/Side coding of band_types higher
than RESERVED_BT (M/S must not be applied to their spectral coefficients,
but marking M/S as present in encode_ms_info() is okay). Much
of the changes here were taken from the decoder and inverted.
This commit does not change the functionality of the decoder as the
previous patch in this series zeroes ms_mask and is_mask.

Reviewed-by: Claudio Freire <klaussfreire@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
libavcodec/aacenc.c

index f7325d4..3e5ac72 100644 (file)
@@ -312,19 +312,26 @@ static void encode_ms_info(PutBitContext *pb, ChannelElement *cpe)
 static void adjust_frame_information(ChannelElement *cpe, int chans)
 {
     int i, w, w2, g, ch;
-    int start, maxsfb, cmaxsfb;
+    int maxsfb, cmaxsfb;
+    IndividualChannelStream *ics;
 
-    for (ch = 0; ch < chans; ch++) {
-        IndividualChannelStream *ics = &cpe->ch[ch].ics;
-        start = 0;
-        maxsfb = 0;
-        cpe->ch[ch].pulse.num_pulse = 0;
+    if (cpe->common_window) {
+        ics = &cpe->ch[0].ics;
         for (w = 0; w < ics->num_windows; w += ics->group_len[w]) {
-            for (w2 = 0; w2 < ics->group_len[w]; w2++) {
-                start = (w+w2) * 128;
+            for (w2 =  0; w2 < ics->group_len[w]; w2++) {
+                int start = (w+w2) * 128;
                 for (g = 0; g < ics->num_swb; g++) {
-                    //apply M/S
-                    if (cpe->common_window && !ch && cpe->ms_mask[w*16 + g]) {
+                    //apply Intensity stereo coeffs transformation
+                    if (cpe->is_mask[w*16 + g]) {
+                        int p = -1 + 2 * (cpe->ch[1].band_type[w*16+g] - 14);
+                        float scale = cpe->ch[0].is_ener[w*16+g];
+                        for (i = 0; i < ics->swb_sizes[g]; i++) {
+                            cpe->ch[0].coeffs[start+i] = (cpe->ch[0].pcoeffs[start+i] + p*cpe->ch[1].pcoeffs[start+i]) * scale;
+                            cpe->ch[1].coeffs[start+i] = 0.0f;
+                        }
+                    } else if (cpe->ms_mask[w*16 + g] &&
+                               cpe->ch[0].band_type[w*16 + g] < NOISE_BT &&
+                               cpe->ch[1].band_type[w*16 + g] < NOISE_BT) {
                         for (i = 0; i < ics->swb_sizes[g]; i++) {
                             cpe->ch[0].coeffs[start+i] = (cpe->ch[0].pcoeffs[start+i] + cpe->ch[1].pcoeffs[start+i]) * 0.5f;
                             cpe->ch[1].coeffs[start+i] = cpe->ch[0].coeffs[start+i] - cpe->ch[1].pcoeffs[start+i];
@@ -332,6 +339,16 @@ static void adjust_frame_information(ChannelElement *cpe, int chans)
                     }
                     start += ics->swb_sizes[g];
                 }
+            }
+        }
+    }
+
+    for (ch = 0; ch < chans; ch++) {
+        IndividualChannelStream *ics = &cpe->ch[ch].ics;
+        maxsfb = 0;
+        cpe->ch[ch].pulse.num_pulse = 0;
+        for (w = 0; w < ics->num_windows; w += ics->group_len[w]) {
+            for (w2 =  0; w2 < ics->group_len[w]; w2++) {
                 for (cmaxsfb = ics->num_swb; cmaxsfb > 0 && cpe->ch[ch].zeroes[w*16+cmaxsfb-1]; cmaxsfb--)
                     ;
                 maxsfb = FFMAX(maxsfb, cmaxsfb);